An integrated approach to enhance the optical performance of couplers based on neural networks, desirability functions and tabu search

Abstract

Resolving a multi-response parameter design problem using the traditional Taguchi method is difficult. This study presents an integrated optimization approach based on neural networks, exponential desirability functions and tabu search to optimize this complicated problem. The proposed approach identifies input control factor settings that maximize the overall minimal satisfaction with all of the responses. The optimal input control factor values are no longer restricted to solution points composed of discrete experimental levels. The proposed approach was implemented for a Taiwanese fiber-optic passive component manufacturer and illustrated through optimizing a fused biconic taper process, thus improving the optical performance of 5% (5/95) single-window broadband couplers. The confirmation results demonstrated the practicability and effectiveness of the proposed approach. This optimization approach was successfully implemented in 5% (5/95) single-window broadband coupler manufacturing and the average defect rate was reduced to about 1.0%, from the previous over 15%.